Spatial models for scaling optimal nutrient management research from plot to field and watershed scales
Issue
Improved technologies for mapping soil nutrient concentrations and corn nitrogen needs will reduce the uncertainty in the 4Rs of fertilizer application, enhance the analysis of environmental conditions that modify soil nutrient concentrations and potentially lead to spatio-temporal models where soil nutrient levels in the growing season can be predicted from soil sampling prior to planting. However, the information learned from research needs to be “scaled up” for applicability to areas beyond the plot scale. This process of extending information to areas that have not been directly studied is essentially an exercise in spatial prediction.
This project will study the interaction of management practices with different soil environments as a basis for understanding how results can transfer to other areas. By identifying and mapping the relationships between natural inherent soil factors and the impact of management practices, researchers aim to develop models capable of predicting soil nutrient outcomes from field-to-watershed scales. The challenge is to account for the factors contributing to soil variation.
Objective
Researchers will analyze how terrain variables induce variation within experimental plots, adding a multi-variate analysis component to an existing project on corn nitrogen needs under cereal rye cover crops. This knowledge will help identify which variables need to be considered in the development of transferable models that can provide a basis for scaling plot-scale research to field- and watershed-scales for optimal nutrient management. As such, this research lays the groundwork to expand development of spatial models to multiple fields with N-rate trials and over multiple years.
Approach
Remote sensing will provide efficient collection of data to serve as covariates for soil properties, such as fine resolution elevation data and multi-temporal, multi-spectral imagery.
Project Updates
Note: Project reports published on the INRC website are often revised from researchers' original reports to increase consistency.
January 2024
The study field was sampled monthly (4-week intervals) from March to October. In each round of sampling, 100 surface samples (15 centimeters or 6 inches) were collected, packed and sent to a commercial laboratory for further fertility analysis. The properties analyzed include nitrate (NO3), phosphorus (P1 and P2), potassium (K), calcium (Ca), magnesium (Mg), pH, buffer pH, cation exchange capacity (CEC) and organic matter. Traditional statistics were calculated for each round of samples to explore the data distribution. The mean, median, standard deviation, and range were included in the analysis. Outliers existing in the data were identified and excluded to create a clean data set, and traditional statistics were also calculated.
The first approach to soil mapping was interpolation through ordinary kriging. Maps were created for each property and each month. These maps were validated using a 90/10 split, meaning that 90% of the data was used for interpolation and 10% to validate the spatial predictions. For each property, geostatistics were calculated to determine further spatial dependence and evaluate the best approach in future analysis (improving map quality). Future analysis will include digital soil mapping with machine learning algorithms and model validation, with experimentation based on the inclusion of drone imagery as potential predictors as well as temporal predictability. Also, the original grid is being divided into the industry standard (2.5-acre grid) to compare results with information currently utilized in corn-soybean production. The same monthly sampling is planned for the upcoming growing season in the same field starting in March 2024.
Samples will be collected and analyzed the same way as the previous season to compare and identify potential trends existing in the data that could be used for temporal predictions.
June 2023
The study field has been sampled monthly (4-week interval) using the grid already created for the initial samples in 2022. In each round of sampling, 100 samples are taken from the field to a depth of 15 cm (6 inches), packed into individual bags and sent for further analysis at a commercial laboratory for soil nutrients.
A stack of covariates, including different digital terrain attributes (slope gradient, profile curvature, eastness, northness, etc.) and remote sensing imagery (Sentinel-2, Landsat-8) has been created to be used as predictors in the modelling process. Digital maps are being created using two approaches, spatial autocorrelation (ordinary kriging) and spatial association (machine learning (ML) with random forest). For the ML method, the covariates are fed into the training process with 80% of the sample results. The spatial prediction power of the respective models is then evaluated against the remaining sample results as an independent validation. The maps obtained from this process are being saved and will later be used to create animations to visualize how soil nutrients vary on a spatial and temporal dimension. Additional analysis is being conducted through statistics of field management zones (e.g., hillslope position) and determination of significant differences and exploring potential connections to landscape processes
December 2022
The study area field was sampled in November as a baseline for soil fertility and soil texture properties. Because this larger suite of soil property analysis required a larger volume of soil to be collected, members from the Miller, Licht and McDaniel labs all worked together to collect the samples on a single day. This was a useful team activity to help everyone become better acquainted and will help with communication as this project continues.
One hundred samples were collected for a sample density of 6.25 samples/ha (0.4-acre grid). These same locations will be sampled through the coming season for comparability and generation of spatial models. Arturo Flores, the master’s student supported by this project, started in January 2023. He will carry out the main soil sampling, spatial modeling and analysis for this project.